DHARMa: Residual Diagnostics for Hierarchical Regression Models

The DHARMa R package provides simulation-based scaled residuals for diagnosing hierarchical regression models. It supports models from packages including 'lme4', 'glmmTMB', 'GLMMadaptive', 'spaMM', 'phylolm', and 'mgcv', and can process simulations from Bayesian software. The package includes plot and test functions for detecting issues like overdispersion, zero-inflation, and autocorrelation.

Use Cases

• Diagnose overdispersion or underdispersion in generalized linear mixed models based on the description of test functions.
• Check for zero-inflation in count data models based on the described model misspecification tests.
• Assess residual spatial, phylogenetic, or temporal autocorrelation based on the package's specialized test functions.
• Validate model fit for Bayesian posterior predictive simulations based on the support for external simulations from 'JAGS', 'STAN', or 'BUGS'.
• Standardize residuals from diverse model classes for intuitive interpretation based on the scaling to values between 0 and 1.

Strengths

• Supports a wide range of model classes from 'lme4', 'glmmTMB', 'GLMMadaptive', 'spaMM', 'phylolm', and 'mgcv'.
• Provides standardized residuals interpretable like linear regression residuals.
• Includes specific tests for common problems like overdispersion, zero-inflation, and autocorrelation.

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Florian Härtig via paperswithcode

Collection Method

Software package for statistical diagnostics.